Reuse fetched PPR neighbor cache entries

[mkolodner-sc]

Moves fetched PPR neighbor rows into the persistent per-state neighbor cache immediately.

Previously, pushResiduals() kept newly fetched neighbor lists in a temporary fetched map for the current push, while older neighbor lists lived in _neighborCache. That meant the push loop had to check both places, and a node that re-entered the frontier later could still trigger another lookup if its neighbor row had not been promoted into _neighborCache.
Now, every fetched (node_id, edge_type_id) row is written to _neighborCache before residuals are pushed. After that, both drainQueue() and the residual-push loop use _neighborCache as the single source of truth for known adjacency.

Example: Example: suppose node u is fetched and pushed in iteration 1. Later, residual flows back to u through another path and u re-enters the frontier. Before this change, u’s adjacency may not have been retained unless it matched the old promotion path. With this change, drainQueue() sees u’s neighbor list in _neighborCache and avoids re-fetching it.

The PPR math is unchanged; this only avoids redundant neighbor fetches and simplifies the push path by removing the split between “newly fetched” and “previously cached” neighbor rows.

[kmontemayor2-sc]

Thanks Matt!